Synergizing CRISPR/Cas9 off-target predictions for ensemble insights and practical applications¶
Shixiong Zhang , Xiangtao Li, Qiuzhen Lin and Ka-Chun Wong
Department of Computer Science, City University of Hong Kong, Kowloon Tong, Hong Kong; Department of Computer Science and Information Technology, Northeast Normal University, Changchun; China and College of Computer Science and Software Engineering, Shenzhen University, Shenzhen.
In [4]:
crop_fig('fpls-07-00506-g001.jpg', 500)
Out[4]:
In [5]:
crop_fig('GS.png', 600)
Out[5]:
Rules of binding/cleaving¶
In [6]:
crop_fig('Fig_Matrices.png', 0)
Out[6]:
In [7]:
get_fig_from_page(1, left=False, full_w=False, top=450, height=1500)
Out[7]:
In [8]:
get_fig_from_page(2, left=True, full_w=True, top=450, height=500)
Out[8]:
Support Vector Machine¶
In [9]:
crop_fig('Fig_SVM.png', 0)
Out[9]:
AdaBoost (Adaptive Boosting)¶
In [10]:
crop_fig('Fig_AdaNoost1.png', 0)
Out[10]:
In [11]:
crop_fig('Fig_Adaboost2.png', 0)
Out[11]:
Decision Tree¶
In [12]:
crop_fig('Fig_DecisionTree.png', 0)
Out[12]:
Random Forest¶
In [13]:
crop_fig('Fig_RandomForest.png', 0)
Out[13]:
Multilayer Perceptron¶
In [22]:
crop_fig('Fig_MLPintro.png', 0)
Out[22]:
In [23]:
crop_fig('Fig_ROC_PRC.png', 0)
Out[23]:
In [15]:
get_fig_from_page(3, left=True, full_w=True, top=450, height=1000)
Out[15]:
In [ ]:
from sklearn.metrics import precision_recall_curve
from sklearn.metrics import average_precision_score
from sklearn.ensemble import AdaBoostClassifier
from sklearn.tree import DecisionTreeClassifier
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from sklearn.model_selection import StratifiedKFold
from sklearn.metrics import roc_curve, auc
from scipy import interp
from itertools import cycle
from sklearn.model_selection import cross_val_score
from sklearn.svm import SVC
from sklearn.neural_network import MLPClassifier
filename = "cdf_score999.csv"
normsocres = np.transpose(np.vstack((MITWeb, MIT, CCTop, Cropit, CFD,
phylop, phastCons, Promoter, Enhancer, segwayPromoter, segwayEnhancer)))
X_train, X_test, y_train, y_test = train_test_split(normsocres, Label, test_size=0.3, random_state=42)
bdt = AdaBoostClassifier(DecisionTreeClassifier(max_depth=1), algorithm="SAMME", n_estimators=200)
rf = RandomForestClassifier(max_depth=1, n_estimators=200)
svc = SVC(kernel = "rbf")
svc1 = SVC(kernel = "sigmoid")
dt = DecisionTreeClassifier(max_depth=6)
mlp = MLPClassifier(alpha=1, activation="relu")
cv = StratifiedKFold(n_splits=5)
In [ ]:
for train, test in cv.split(sub_normsocres, Label):
scores = bdt.fit(sub_normsocres[train], Label[train]).decision_function(sub_normsocres[test])
# Compute ROC curve and area the curve
fpr, tpr, thresholds = roc_curve(Label[test], scores)
tprs.append(interp(mean_fpr, fpr, tpr))
tprs[-1][0] = 0.0
roc_auc = auc(fpr, tpr)
aucs.append(roc_auc)
plt.plot(fpr, tpr, lw=1, alpha=0.3, label='ROC fold %d (AUC = %0.4f)' % (i, roc_auc))
i += 1
In [16]:
get_fig_from_page(4, left=True, full_w=True, top=450, height=2000)
Out[16]:
In [17]:
get_fig_from_page(4, left=True, full_w=True, top=2350, height=2000)
Out[17]:
In [18]:
get_fig_from_page(4, left=True, full_w=True, top=4350, height=2000)
Out[18]:
In [19]:
get_fig_from_page(5, left=True, full_w=True, top=450, height=1800)
Out[19]:
In [ ]:
for train, test in cv.split(sub_normsocres, Label):
scores = bdt.fit(sub_normsocres[train], Label[train]).decision_function(sub_normsocres[test])
#scores = svc.fit(sub_normsocres[train], Label[train]).decision_function(sub_normsocres[test])
#scores = svc1.fit(sub_normsocres[train], Label[train]).decision_function(sub_normsocres[test])
#scores = dt.fit(sub_normsocres[train], Label[train]).predict_proba(sub_normsocres[test])[:, 1]
#scores = rf.fit(sub_normsocres[train], Label[train]).predict_proba(sub_normsocres[test])[:, 1]
#scores = mlp.fit(sub_normsocres[train], Label[train]).predict_proba(sub_normsocres[test])[:, 1]
# Compute ROC curve and area the curve
fpr, tpr, thresholds = roc_curve(Label[test], scores)
tprs.append(interp(mean_fpr, fpr, tpr))
tprs[-1][0] = 0.0
roc_auc = auc(fpr, tpr)
aucs.append(roc_auc)
plt.plot(fpr, tpr, lw=1, alpha=0.3, label='ROC fold %d (AUC = %0.4f)' % (i, roc_auc))
i += 1
mean_tpr = np.mean(tprs, axis=0)
mean_tpr[-1] = 1.0
mean_auc = auc(mean_fpr, mean_tpr)
std_auc = np.std(aucs)
plt.plot(mean_fpr, mean_tpr, color='b',
label=r'Mean ROC (AUC = %0.4f $\pm$ %0.2f)' % (mean_auc, std_auc),
lw=2, alpha=.8)
std_tpr = np.std(tprs, axis=0)
tprs_upper = np.minimum(mean_tpr + std_tpr, 1)
tprs_lower = np.maximum(mean_tpr - std_tpr, 0)
In [ ]:
plt.figure(figsize=(7, 7))
LabelAda = "AdaBoost, AUC %.4f" % meanaucAda
LabelRF = "RandomForest, AUC %.4f" % meanaucRF
LabelDT = "DecisionTree, AUC %.4f" % meanaucDT
LabelSVC = "SVC, AUC %.4f" % meanaucsvc
LabelMLP = "MLP, AUC %.4f" % meanaucMLP
plt.plot(mean_fpr, meanAda, color='r', label= LabelAda, ls = "--", marker ='*')
plt.plot(mean_fpr, meanMLP, color='grey', label= LabelMLP, ls = ":")
plt.plot(mean_fpr, meanRF, color='darkblue', label= LabelRF, ls = "--")
plt.plot(mean_fpr, meanDT, color='green', label= LabelDT, ls = "-.")
plt.plot(mean_fpr, meansvc, color='b', label= LabelSVC, ls = ":", marker ='.')
In [30]:
get_fig_from_page(5, left=False, full_w=False, top=2250, height=4000, shrink=0.15)
Out[30]:
Case study¶
In [21]:
get_fig_from_page(6, left=True, full_w=True, top=450, height=1800)
Out[21]:
In [35]:
crop_fig('20171113_8gRNAs_GUIDEseq_CAS_3K.png', 1200)
Out[35]:
Discussion¶
What other benchmark sutdies we could do to improve the ensemble models?¶
What other machine learning algorithm we could use to compare with the proposed ensemble model in this study?¶
Brainstorm: How would you design a better method to get the "rule of binding/cleaving" for CRISPR/Cas9 that outcompetes CFD matrix?¶
In [ ]: